Due to high densification, high performance and other factors of servers, an amount of power consumed in a data center has been increasing, and the increase of the power consumption has led to an increase in heat generation. Temperature rise due to an increase in heat generation causes a thermal runaway or a failure of the server. This makes it difficult to operate the data center stably, and brings about adverse effects such as reduction of the service life of the server.
To solve the above problems, the air conditioner is provided in a place such as a server room where servers are arranged. However, high temperature air may stagnate in an area not easily reachable by a flow of air generated by the air conditioner (such an area is hereinafter referred to as a hot spot). This phenomenon is described with reference to FIG. 1. In the example of FIG. 1, servers A to P and an air conditioner are installed in a server room. The air conditioner lowers the temperature in the server room by sending air. The servers A to P execute jobs.
Since the servers A to H arranged in a location relatively close to the air conditioner are cooled by a strong air flow sent from the air conditioner, the hot spot hardly occurs around the servers A to H. However, the servers I to P arranged in a location relatively far from the air conditioner are hardly cooled since the wind sent from the air conditioner is weak. In the farthest location from the air conditioner, in particular, within the surroundings of the servers I to P, a hot spot is likely to occur because the air from the air conditioner is not easily reachable.
In this way, how easily the temperature rises varies among places where servers are installed. Also, how easily the temperature rises varies due to a difference in the processing load of resident basic software and so on. Since a temperature rise poses various problems as described above, jobs are preferably allocated so that a local temperature rise will not occur. However, even if a temperature rise is detected after the execution of a job is started, it is ill-advised to terminate the job easily, because the terminating of the ongoing job wastes an already executed portion of the job.
There is a technique of allocating tasks and the like based on temperature information. For example, a document discloses a technique that measures the temperature of each of multiple processors, and allocates tasks to the multiple processors based on the measured temperatures.
However, even a method of allocating tasks to processors currently at low temperature, for example, may cause a significant temperature rise, if a task of a high processing load is allocated to a processor in which the temperature tends to rise easily.
As an example of prior arts, there are known International Publication Pamphlet No. 2003/083693, Japanese Laid-open Patent Publication No. 2006-133995, Japanese Laid-open Patent Publication No. 2008-242614, and Japanese Laid-open Patent Publication No. 2010-108324.